Neural Architecture Search in Graph Neural Networks

TL;DR

This paper compares reinforcement learning and evolutionary algorithm-based Neural Architecture Search methods for optimizing Graph Neural Networks, revealing that both methods perform similarly to random search across multiple datasets and search spaces.

Contribution

It provides a comparative analysis of NAS techniques for GNNs, highlighting the limited impact of current search strategies on performance.

Findings

Both NAS methods achieve similar accuracy to random search.

Many search space dimensions may be irrelevant to GNN performance.

Current NAS approaches may not significantly outperform naive search methods.

Abstract

Performing analytical tasks over graph data has become increasingly interesting due to the ubiquity and large availability of relational information. However, unlike images or sentences, there is no notion of sequence in networks. Nodes (and edges) follow no absolute order, and it is hard for traditional machine learning (ML) algorithms to recognize a pattern and generalize their predictions on this type of data. Graph Neural Networks (GNN) successfully tackled this problem. They became popular after the generalization of the convolution concept to the graph domain. However, they possess a large number of hyperparameters and their design and optimization is currently hand-made, based on heuristics or empirical intuition. Neural Architecture Search (NAS) methods appear as an interesting solution to this problem. In this direction, this paper compares two NAS methods for optimizing GNN: one based on reinforcement learning and a second based on evolutionary algorithms. Results consider 7 datasets over two search spaces and show that both methods obtain similar accuracies to a random search, raising the question of how many of the search space dimensions are actually relevant to the problem.